Generalisation and domain specific functions in genetic programming

This research presents an evaluation of user defined domain specific functions of genetic programming using relational learning problems, generalisation for this class of learning problems and learning bias. After providing a brief theoretical background, two sets of experiments are detailed: experiments and results concerning the Monk-2 problem and experiments attempting to evolve generalising solutions to parity problems with incomplete data sets. The results suggest that using non-problem specific functions may result in greater generalisation for relational problems